How I Use Hermes Agent as My Always-On Research and Writing Assistant

My setup for an 8am daily briefing, 10pm daily and weekly summaries, paper writing, blogging, and remote work from my phone.

Gordon Guocheng Qian

May 18, 2026 · 7 min read

I have been using Hermes Agent as a personal AI operating layer around my research, writing, and daily workflow.

The important part is not that it can chat. Many tools can chat. The useful part is that Hermes can run as a persistent agent with memory, tools, scheduled jobs, file access, web access, and messaging-platform delivery. That turns it from a chatbot into something closer to a personal research assistant that can work in the background and report back when needed.

My current setup has three main loops:

An 8am daily briefing for research, market watch, and todos.
A 10pm daily summary of what I worked on and what should carry over.
A weekly summary that turns the week into a higher-level work log.

Phone interface controlling Hermes Agent on a Mac mini with access to local files — From my phone, I can ask Hermes to inspect a local repo, summarize recent work, polish a paper section, prepare a blog draft, run a quick terminal check, or leave a report for when I return to the Mac mini.

The basic setup

Hermes runs locally on my Mac mini and connects to my tools through a terminal-first workflow. Because the agent lives on the machine that holds my local repositories, paper drafts, notes, and blog files, a phone message can become real work on the filesystem rather than just another chat reply. The rough architecture is:

Mac / local filesystem
        ↓
Hermes Agent CLI + tool system
        ↓
Web search, file editing, terminal commands, memory, session search, cron jobs
        ↓
Feishu/Lark delivery to my phone

I use the terminal version for deep work, especially paper editing and repo-level tasks. I also connect Hermes to Feishu/Lark so that scheduled briefings and summaries can be delivered directly to my phone.

The setup is intentionally boring: local files stay local, scheduled jobs run through Hermes cron, and messages arrive through a normal chat app. That is what makes it practical.

Scheduled jobs

The recurring automation is handled with Hermes cron jobs. Conceptually, it looks like this:

8:00am  daily briefing
10:00pm daily work summary
10:00pm weekly summary / weekly reflection checkpoint

The 8am job uses web and research tools. The 10pm jobs use session search and memory-aware summarization, so Hermes can look back over the actual work I did rather than asking me to reconstruct the day manually.

In practice, the cron prompts matter more than the schedule itself. The prompts specify what kind of information I care about, how compact the output should be, and which work categories must not be missed.

8am: daily briefing

Every morning, I want a short but useful overview before I start the day.

My daily briefing combines several things:

Top AI reads: papers, technical reports, model releases, and important blog posts.
Research relevance: why a paper or release matters for video generation, VLMs, diffusion/flow models, multimodal agents, or 3D generation.
Market watch: high-level movement in AI-related companies and sectors I care about.
Todos: a small list of concrete things I should consider doing today.

The key is that the briefing is not a generic news digest. It is tuned to my research taste. For example, an ordinary AI newsletter may mention a new model release. Hermes tries to answer the more useful question: does this change what I should read, write, test, or think about?

I also like the paper section to include direct source links, especially arXiv abstract links. That makes the briefing actionable: I can open the paper immediately from my phone and decide whether it deserves deeper reading later.

10pm: daily and weekly summaries

The 10pm summary is the opposite of the 8am briefing. In the morning, Hermes looks outward. At night, it looks backward.

The daily summary answers:

What did I actually work on today?
Which paper edits, manuscript changes, code changes, or blog drafts mattered?
What decisions did I make?
What should continue tomorrow?

This is useful because a lot of research work is fragmented. A day might include one paper section rewrite, one figure polish, one related-work search, one code experiment, and one blog draft. Without a summary, the day feels scattered. With a summary, the work becomes a trace.

The weekly summary is a higher-level version of the same idea. It compresses daily activity into themes: papers advanced, submissions prepared, experiments debugged, writing completed, and pending follow-ups.

The goal is not journaling for its own sake. The goal is continuity. Hermes helps turn many small interactions into a persistent work memory.

Paper writing and polishing

The most valuable use case for me is paper writing.

Hermes can work directly with local paper repositories. That means it can inspect LaTeX files, edit sections, polish paragraphs, check consistency, and run build or search commands when needed.

I use it for several kinds of paper work:

Grammar and clarity polishing: clean up language without changing the technical meaning.
Related work organization: compare papers, group citations, and improve narrative flow.
Method-section rewriting: make the structure clearer while preserving technical claims.
Abstract and introduction iteration: sharpen the motivation, contribution, and positioning.
Submission preparation: check files, captions, references, and paper-specific TODOs.

The important thing is that Hermes remembers my preferences. For example, when I ask for a grammar polish, I usually do not want it to rewrite the whole argument. I want it to fix grammar, typos, and awkward phrasing while preserving my intent.

This is where persistent memory and skills become useful. Over time, Hermes learns the difference between “rewrite this section” and “lightly polish this paragraph.”

Blogging and technical writing

I also use Hermes for blog writing.

My blogging workflow is usually:

I write rough notes or a thesis.
Hermes turns them into a structured draft.
I ask for grammar polishing or stronger flow.
Hermes creates a clean local HTML or Markdown version.
I manually review the final version before posting.

This is useful for speculative technical posts, where the goal is not just correctness but also positioning: what is the thesis, what is the ranking of possibilities, what evidence is public, and where should the claims be hedged?

Hermes is especially good at helping me keep a blog post in the right mode. Some posts should be precise and technical. Some should be conversational. Some should be formatted for X or LinkedIn. Some should remain as a clean standalone HTML file.

Remote work from my phone

The mobile workflow is what makes the system feel different from a normal terminal agent.

Because Hermes can deliver through Feishu/Lark, I can interact with it from my phone. That means I can:

receive the 8am briefing without opening my laptop;
check the 10pm summary while away from my desk;
send a quick instruction when I remember something on the go;
ask it to inspect or prepare local work before I return to my computer;
use my phone as a lightweight command center for research and writing tasks.

This does not replace deep work on a laptop. It removes friction around it. If I have an idea while walking, commuting, or between meetings, I can send it to Hermes. If a scheduled summary reminds me of a missing TODO, I can reply immediately. The agent becomes a bridge between mobile moments and desktop execution.

The practical result is that my research workflow is less tied to one physical context. The Mac still does the heavy lifting, but the phone becomes a remote interface to the same assistant.

Supervised FSD commute with a phone connected to a remote AI agent on a Mac mini — In the supervised-driving version of the same loop, FSD lowers commute friction while Hermes keeps work moving in the background. FSD + Agent = productivity: travel time becomes a lightweight control surface for research, writing, and file-based tasks.

Why this matters

The biggest benefit is continuity.

Research and writing are not single-turn tasks. They unfold across days: reading papers, forming opinions, rewriting sections, updating claims, preparing submissions, and remembering what changed yesterday.

Hermes helps connect those fragments. The daily briefing brings new information in. The nightly summary compresses the day. The weekly summary turns scattered progress into a higher-level record. The writing workflows keep paper and blog drafts moving. The mobile interface keeps the loop alive even when I am not at my desk.

That is why I think of Hermes less as a chatbot and more as an always-on research and writing assistant.